<p>The application of machine learning (ML) in the simulation and optimization of solar cells has garnered significant attention due to its potential to accelerate design processes, improve efficiency, and reduce computational costs. A BaZrS₃ based chalcogenide perovskite solar cell having an n-i-p architecture was utilised in this investigation. SCAPS-1D program was utilised to generate a total of 2000 data points by varying two input parameters simultaneously for 5 different input–output combinations. This paper presents a novel Hybrid Regression model leveraging the strengths of both Random Forest (RF) and XGBoost, two popular decision tree ensemble methods to predict the output parameters of solar cell for different input–output combinations. The model aims to achieve optimal predictive performance by balancing low bias and moderate variance, a critical consideration in regression tasks. This balance is maintained through dynamically adjusted weights that adapt to different input feature combinations. The proposed Hybrid model shows better performance (in both training and inference) compared to custom shallow network and TabNet, as measured by Mean Absolute Error (MAE) and Root Mean Square Error (RMSE). Thorough hyper parameter optimization, dynamic weighting, and rigorous cross-validation across diverse test scenarios ensure the model is fully fine-tuned and establish it as a resource-efficient and reliable alternative to more complex deep learning approaches. The close agreement between the SCAPS simulation and the proposed model is demonstrated by the negligible differences in their results. Specifically, the percentage errors for V_OC, J_SC, FF, and PCE are remarkably low, measuring 0.12%, 0.16%, 0.18%, and 0.28%, respectively.</p>